[Freeswitch-dev] core odbc vs in-module mysql
miconda at gmail.com
Fri Feb 26 15:59:19 PST 2010
On 02/26/2010 08:46 PM, Michael Jerris wrote:
> lots of people complain about the huge overhead of odbc and how horrible it is,
there are lot of discussions on the web about this topic, apart of
casual fud cases, perhaps there is something generating all these.
However, personally I haven't complained, just wanted to know if the FS
core API around ODBC gives some special goodies (e.g., reconnect,
auto/nicer handling of error cases, ...).
> and every time I ask them to show me some real numbers that show a significant performance impact.
It is not only about performance, but also losing (or adding overhead to
access) particular extensions of the sql engine behind. E.g., retrieving
the last inserted id (auto-increment) requires a second query, with
error handling a.s.o. while the native lib provides a functions.
> Of all those who have loudly complained about the horrors of odbc, I have not had a single person take me up on my request and show me any numbers at all to support this theory.
Do you you mean generic myodbc vs libmyqlclient performance figures or
for FS specific needs? Not a db expert but I think the difference is
revealed better by complex queries rather than simple
> If there is not any real numbers to show a real impact, we can not make a call about if the performance difference would justify increased complexity of supporting anything other than odbc.
Again, it was not about changing something, but collecting hints about
what to use in a custom extension.
> On Feb 26, 2010, at 1:06 PM, Daniel-Constantin Mierla wrote:
>> On 02/26/2010 05:08 PM, Anthony Minessale wrote:
>>> We use the odbc for any in-tree sql stuff because we are trying to remain database agnostic.
>>> if it's your own module it's ok but if you want to contribute it, we would appreciate using ODBC
>> ok, thanks. If I understood the architecture, default sql engine in the core is sqlite, with option to replace it with odbc at compile time.
>> odbc is indeed an abstract layer, but you lose from the power of underneath sql engine. And some out there don't like odbc much, asking to avoid it :-) .
> Sqlite is mostly there just so it works out of the box, any serious implementation should probably use something via odbc.
> FreeSWITCH-dev mailing list
> FreeSWITCH-dev at lists.freeswitch.org
Kamailio SIP Router Masterclass, Berlin, March 22-26, 2010
More information about the FreeSWITCH-dev